仿生式自生长自组织神经网络探索

Exploration of bionic self-growing and self-organizing neuron network

为探索仿生式可生长型人工神经网络的特点和优势,提出一套仿生式、可自生长神经网络的生成与组织算法。采用数据并行处理的方式进行神经网络计算,加快大量大规模网络的生成和功能验证与筛选;完成以现场可编程门阵列的架构为参考的软件仿真平台的设计,基于此仿真平台进行算法验证。实验结果表明,基于该算法生成的网络,随着输入信号的增加,具有自生长和自组织特性,通过筛选机制,能够生成可识别不同输入的有效网络。

For exploring features and advantages of the bionic growing neural network, an algorithm to build self-growing and self-organizing neuron network was presented. A parallel processing method was presented to enhance the computation efficiency of the presented algorithm, and to help building large scale of neuron network with reasonable time. To verify the proposed algo- rithm, a software simulation platform referred to the architecture of FPGA was implemented. Results show that, the neuron network built using the above algorithm can self-grow and self-organize as the complexity of the input external signals increases. And with the screening mechanism, neuron network that can identify different input external signals is built successfully.